In Project 540 "Use of Artificial Intelligence in Medical Diagnostic and Prognostic Systems" the Fraunhofer Heinrich Hertz Institute developed prototypical quality characteristics for AI methods and training data in medical diagnostics on behalf of the BSI, using a use case in ECG diagnostics. The application of artificial intelligence in medical diagnostics can significantly improve patient care, especially through the early detection of pathologies and a potentially improved treatment outcome. The safety, explainability, and performance of AI methods must be demonstrably ensured for both patients and users to enable safe and efficient operation and to strengthen trust in AI-supported medical diagnostics. Given the likely classification as a high-risk application under the European AI Regulation, particularly high requirements must be met for AI methods, while no generally used testing tools are currently available.

This project, conceived as a feasibility study, examines a possible architecture of quality characteristics and validates them through feedback from medical personnel, initially discussing the architecture of existing AI methods and training data in the context of ECG diagnostics. Subsequently, quality characteristics are developed, and requirements for them in terms of safety, explainability, and performance are created in the form of requirement profiles. The suitability of the AI methods is finally tested and evaluated based on the requirement profiles, and the requirement profiles themselves are adjusted through clinical and technical feedback.